Notes for Team 6

(From September 25 meeting with Prof. Keshav)

Remarks

• This team is responsible for the integration part (and for "cracking the whip"). The first task will be to get each team to check in a null interface.
• Then there is the whole area of protocol-specific management, a subject of on-going research, especially project specification management. Commercial systems such as Tivoli and Unicenter have databases which specify what runs on each system being managed.
• All network managers need application specific knowledge of what is going on, which is more than only the basic SNMP stuff. For example, the whiteboard application might want to know who is where.
• Helping the network manager to analyze the managed system is the MIB (Management Information Database). This is complex, but hierarchical, which is nice.
• You can ask other teams what information they are willing to give you, and then you have to figure out how to implement the management system in a nice way. Then you write a stub for the other teams showing what fields you want them to fill in and how. You could design your stubs to listen on a separate port for a manager request. The goal is to design management into every layer from the start.
• Alarms - asynchronous alarms from managed objects go to the System Manager.
• Authentication - SNMP has the notion of community: who can access what objects. "Public" is the least restrictive. From there you might go to passwords required. (Passwords in clear text are OK for this project.) Or use challenge/response, which is one step better.
• Your major challenge is to figure out what the managed objects should be.

Questions

1. Is SourceSave available in CSUGLAB? It should be. If not, send Prof. Keshav email.
2. Does the Manager have any active control over other layers? Can the Manager drop a phone call? That's the next step. If you get the passive management down, you could go on to implement this but it requires major security considerations.
3. Shouldn`t there be security anyway? Usually we just use the idea of "trusted group". For a book on SNMP, see "The simple book: an introduction to networking management", M.T. Rose, in Carpenter Library.
4. The stub sits on the client? Yes.
5. Isn't that just the same as SNMP? Yes, but in SNMP all the information in the hook is stored in a tree, which you can then interrogate. This might be slightly heavyweight for the project.
6. Do we have to use SNMP? No, but you should understand it.
7. Could there be several manager servers? Yes, each would register themself first, perhaps via the directory service. For example, there could be many gateways. Check for timeouts to determine livliness. For consistency, whenever you save state, figure out when it will be removed.
8. Do we set up the whole interface, or let the teams do it? The latter. First cut is due September 30.
9. How are stubs invoked? Listen on a port for a request, send a packet in some format. Or you could use remote invocation, which is not as simple. Or you could just incorporate SNMP, but that is a lot of stuff.